Add GPU Faust fancy idx support in pyfaust wrapper.

Refactor fancy idx functions from TransformHelper<FPP,Cpu> into parent TransformHelperGen,
Update Transform(Helper)<FPP,GPU2> for fancy idx support.

src/faust_linear_operator/CPU/faust_TransformHelper.h

@@ -80,7 +80,6 @@ namespace Faust
 #ifdef USE_GPU_MOD
 			FaustGPU<FPP> *gpu_faust;
 #endif
-			void eval_fancy_idx_Transform();
 			public:
 			TransformHelper(const std::vector<MatGeneric<FPP,Cpu> *>& facts, const FPP lambda_ = (FPP)1.0, const bool optimizedCopy=false, const bool cloning_fact = true, const bool internal_call=false);
 			TransformHelper();
@@ -148,7 +147,6 @@ namespace Faust
 					faust_unsigned_int* num_rows,
 					faust_unsigned_int* num_cols,
 					const bool transpose = false) const;
-			TransformHelper<FPP, Cpu>* fancy_index(faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols);
 			MatDense<FPP,Cpu> get_product();// const;
 			void get_product(MatDense<FPP,Cpu>& prod) const;
 			void save_mat_file(const char* filename) const;

src/faust_linear_operator/CPU/faust_TransformHelper.hpp

@@ -137,30 +137,7 @@ namespace Faust {
 	template<typename FPP>
 		TransformHelper<FPP,Cpu>::TransformHelper(TransformHelper<FPP,Cpu>* th, faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols): TransformHelper<FPP,Cpu>()
 	{
-		this->transform = th->transform; //do not remove this line, necessary for eval*()
-		this->copy_transconj_state(th);
-		this->is_sliced = false;
-		//TODO: check indices
-		//				handleError("Faust::TransformHelper::TransformHelper(TransformHelper,Slice)", "Fancy indexing overflows a Faust dimension.");
-		unsigned int id0=0, id1=1;
-		this->fancy_num_cols = num_cols;
-		this->fancy_num_rows = num_rows;
-		if(this->is_transposed)
-		{
-			id0 = 1;
-			id1 = 0;
-			this->fancy_num_cols = num_rows;
-			this->fancy_num_rows = num_cols;
-		}
-		this->fancy_indices[id0] = new faust_unsigned_int[num_rows];
-		this->fancy_indices[id1] = new faust_unsigned_int[num_cols];
-		this->is_fancy_indexed= true;
-		memcpy(this->fancy_indices[id0], row_ids, num_rows*sizeof(faust_unsigned_int));
-		memcpy(this->fancy_indices[id1], col_ids, num_cols*sizeof(faust_unsigned_int));
-		eval_fancy_idx_Transform();
-		delete[] this->fancy_indices[0];
-		delete[] this->fancy_indices[1];
-		copy_mul_mode_state(th);
+		this->init_fancy_idx_transform(th, row_ids, num_rows, col_ids, num_cols);
 	}
 
 	template<typename FPP>
@@ -931,58 +908,6 @@ namespace Faust {
 				Faust::conjugate(elts,*num_cols*(*num_rows));
 		}
 
-	template<typename FPP>
-		TransformHelper<FPP, Cpu>* TransformHelper<FPP,Cpu>::fancy_index(faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols)
-		{
-			return new TransformHelper<FPP,Cpu>(this, row_ids, num_rows, col_ids, num_cols);
-		}
-
-	template<typename FPP>
-		void TransformHelper<FPP, Cpu>::eval_fancy_idx_Transform()
-		{
-			bool cloning_fact = false;
-			faust_unsigned_int size = this->size();
-			std::vector<MatGeneric<FPP,Cpu>*> factors((size_t) size);
-			MatGeneric<FPP,Cpu>* gen_fac, *first_sub_fac, *last_sub_fac;
-			gen_fac = this->transform->get_fact(0, cloning_fact);
-			//				first_sub_fac = gen_fac->get_rows(this->slices[0].start_id, this->slices[0].end_id-this->slices[0].start_id);
-			//		first_sub_fac->Display();
-			first_sub_fac = gen_fac->get_rows(this->fancy_indices[0], this->fancy_num_rows);
-			if(cloning_fact)
-				delete gen_fac;
-			if(size > 1) {
-				gen_fac = this->transform->get_fact(size-1, cloning_fact);
-				//					last_sub_fac = gen_fac->get_cols(this->slices[1].start_id, this->slices[1].end_id-this->slices[1].start_id);
-				last_sub_fac = gen_fac->get_cols(this->fancy_indices[1], this->fancy_num_cols);	//		std::cout << "---" << std::endl;
-				//		last_sub_fac->Display();
-				if(cloning_fact)
-					delete gen_fac;
-				factors.reserve(size);
-				factors.insert(factors.begin(), first_sub_fac);
-				if(size > 2)
-				{
-					auto it = factors.begin();
-					for(faust_unsigned_int i = 1; i < size-1; i++)
-					{
-						gen_fac = this->transform->get_fact(i, cloning_fact);
-						factors[i] = gen_fac;
-					}
-				}
-				factors.insert(factors.begin()+(size-1), last_sub_fac);
-				factors.resize(size);
-			}
-			else { //only one factor
-				last_sub_fac = first_sub_fac->get_cols(this->fancy_indices[1], this->fancy_num_cols);
-				delete first_sub_fac;
-				factors[0] = last_sub_fac;
-				factors.resize(1);
-			}
-			this->transform = make_shared<Transform<FPP, Cpu>>(factors, 1.0, false, cloning_fact);
-			if(cloning_fact) {
-				for(faust_unsigned_int i = 0; i < size; i++)
-					delete factors[i];
-			}
-		}
 
 
 

src/faust_linear_operator/GPU2/faust_TransformHelper_gpu.h

@@ -15,6 +15,7 @@ namespace Faust
 			public:
 				TransformHelper();
 				TransformHelper(TransformHelper<FPP,GPU2>* th, Slice s[2]);
+				TransformHelper(TransformHelper<FPP,GPU2>* th, faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols);
 				TransformHelper(const std::vector<MatGeneric<FPP,GPU2> *>& facts, const FPP lambda_ = (FPP)1.0, const bool optimizedCopy=false, const bool cloning_fact = true, const bool internal_call=false);
 				TransformHelper(const TransformHelper<FPP,Cpu>& cpu_t, int32_t dev_id=-1, void* stream=nullptr);
                 TransformHelper(const TransformHelper<FPP,GPU2>& th, bool transpose, bool conjugate);
@@ -67,10 +68,6 @@ namespace Faust
 				TransformHelper<FPP,GPU2>* transpose();
 				TransformHelper<FPP,GPU2>* conjugate();
 				TransformHelper<FPP,GPU2>* adjoint();
-//				TransformHelper<FPP, GPU2>* slice(faust_unsigned_int start_row_id, faust_unsigned_int end_row_id,
-//						faust_unsigned_int start_col_id, faust_unsigned_int end_col_id);
-				TransformHelper<FPP, GPU2>* fancy_index(faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols);
-
 				TransformHelper<FPP,GPU2>* optimize_time(const bool transp=false, const bool inplace=false, const int nsamples=1);
 				TransformHelper<FPP,GPU2>* optimize(const bool transp=false);
 				TransformHelper<FPP,GPU2>* optimize_storage(const bool time=true);

src/faust_linear_operator/GPU2/faust_TransformHelper_gpu.hpp

@@ -38,6 +38,12 @@ namespace Faust
 		this->init_sliced_transform(th, s);
 	}
 
+	template<typename FPP>
+		TransformHelper<FPP,GPU2>::TransformHelper(TransformHelper<FPP,GPU2>* th, faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols): TransformHelper<FPP,GPU2>()
+	{
+		this->init_fancy_idx_transform(th, row_ids, num_rows, col_ids, num_cols);
+	}
+
 #ifndef IGNORE_TRANSFORM_HELPER_VARIADIC_TPL
 	template<typename FPP>
 		template<typename ... GList>
@@ -455,19 +461,6 @@ namespace Faust
             return t;
 		}
 
-//	template<typename FPP>
-//		TransformHelper<FPP, GPU2>* TransformHelper<FPP, GPU2>::slice(faust_unsigned_int start_row_id, faust_unsigned_int end_row_id,
-//				faust_unsigned_int start_col_id, faust_unsigned_int end_col_id)
-//		{
-//		}
-
-	template<typename FPP>
-		TransformHelper<FPP, GPU2>* TransformHelper<FPP,GPU2>::fancy_index(faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols)
-		{
-			throw std::runtime_error("fancy_index is yet to implement in Faust C++ core for GPU.");
-			return nullptr;
-		}
-
 	template<typename FPP>
 		faust_unsigned_int TransformHelper<FPP,GPU2>::getNBytes() const
 		{

src/faust_linear_operator/faust_TransformHelperGen.h

@@ -88,10 +88,12 @@ namespace Faust
 			int get_mul_order_opt_mode() const;
 			int get_Fv_mul_mode() const;
 			void eval_sliced_Transform();
+			void eval_fancy_idx_Transform();
 			TransformHelper<FPP, DEV>* slice(faust_unsigned_int start_row_id, faust_unsigned_int end_row_id,
 					faust_unsigned_int start_col_id, faust_unsigned_int end_col_id);
-
+			TransformHelper<FPP, DEV>* fancy_index(faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols);
 		protected:
+			void init_fancy_idx_transform(TransformHelper<FPP,DEV>* th, faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols);
 			void init_sliced_transform(TransformHelper<FPP,DEV>* th, Slice s[2]);
 			bool is_transposed;
 			bool is_conjugate;

src/faust_linear_operator/faust_TransformHelperGen.hpp

@@ -15,6 +15,36 @@ namespace Faust
 		this->is_conjugate = conjugate?!th->is_conjugate:th->is_conjugate;
     }
 
+
+	template<typename FPP, FDevice DEV>
+	void TransformHelperGen<FPP,DEV>::init_fancy_idx_transform(TransformHelper<FPP,DEV>* th, faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols)
+	{
+		this->transform = th->transform; //do not remove this line, necessary for eval*()
+		this->copy_transconj_state(*th);
+		this->is_sliced = false;
+		//TODO: check indices
+		//				handleError("Faust::TransformHelper::TransformHelper(TransformHelper,Slice)", "Fancy indexing overflows a Faust dimension.");
+		unsigned int id0=0, id1=1;
+		this->fancy_num_cols = num_cols;
+		this->fancy_num_rows = num_rows;
+		if(this->is_transposed)
+		{
+			id0 = 1;
+			id1 = 0;
+			this->fancy_num_cols = num_rows;
+			this->fancy_num_rows = num_cols;
+		}
+		this->fancy_indices[id0] = new faust_unsigned_int[num_rows];
+		this->fancy_indices[id1] = new faust_unsigned_int[num_cols];
+		this->is_fancy_indexed = true;
+		memcpy(this->fancy_indices[id0], row_ids, num_rows*sizeof(faust_unsigned_int));
+		memcpy(this->fancy_indices[id1], col_ids, num_cols*sizeof(faust_unsigned_int));
+		this->eval_fancy_idx_Transform();
+		delete[] this->fancy_indices[0];
+		delete[] this->fancy_indices[1];
+		copy_mul_mode_state(*th);
+	}
+
 	template<typename FPP, FDevice DEV>
 		void TransformHelperGen<FPP,DEV>::init_sliced_transform(TransformHelper<FPP,DEV>* th, Slice s[2])
 
@@ -313,4 +343,57 @@ namespace Faust
 			return new TransformHelper<FPP, DEV>(dynamic_cast<TransformHelper<FPP, DEV>*>(this), s);
 		}
 
+	template<typename FPP, FDevice DEV>
+		TransformHelper<FPP, DEV>* TransformHelperGen<FPP,DEV>::fancy_index(faust_unsigned_int* row_ids, faust_unsigned_int num_rows, faust_unsigned_int* col_ids, faust_unsigned_int num_cols)
+		{
+			return new TransformHelper<FPP,DEV>(dynamic_cast<TransformHelper<FPP, DEV>*>(this), row_ids, num_rows, col_ids, num_cols);
+		}
+
+	template<typename FPP, FDevice DEV>
+		void TransformHelperGen<FPP, DEV>::eval_fancy_idx_Transform()
+		{
+			bool cloning_fact = false;
+			faust_unsigned_int size = this->size();
+			std::vector<MatGeneric<FPP,DEV>*> factors((size_t) size);
+			MatGeneric<FPP,DEV>* gen_fac, *first_sub_fac, *last_sub_fac;
+			gen_fac = this->transform->get_fact(0, cloning_fact);
+			//				first_sub_fac = gen_fac->get_rows(this->slices[0].start_id, this->slices[0].end_id-this->slices[0].start_id);
+			//		first_sub_fac->Display();
+			first_sub_fac = gen_fac->get_rows(this->fancy_indices[0], this->fancy_num_rows);
+			if(cloning_fact)
+				delete gen_fac;
+			if(size > 1) {
+				gen_fac = this->transform->get_fact(size-1, cloning_fact);
+				//					last_sub_fac = gen_fac->get_cols(this->slices[1].start_id, this->slices[1].end_id-this->slices[1].start_id);
+				last_sub_fac = gen_fac->get_cols(this->fancy_indices[1], this->fancy_num_cols);	//		std::cout << "---" << std::endl;
+				//		last_sub_fac->Display();
+				if(cloning_fact)
+					delete gen_fac;
+				factors.reserve(size);
+				factors.insert(factors.begin(), first_sub_fac);
+				if(size > 2)
+				{
+					auto it = factors.begin();
+					for(faust_unsigned_int i = 1; i < size-1; i++)
+					{
+						gen_fac = this->transform->get_fact(i, cloning_fact);
+						factors[i] = gen_fac;
+					}
+				}
+				factors.insert(factors.begin()+(size-1), last_sub_fac);
+				factors.resize(size);
+			}
+			else { //only one factor
+				last_sub_fac = first_sub_fac->get_cols(this->fancy_indices[1], this->fancy_num_cols);
+				delete first_sub_fac;
+				factors[0] = last_sub_fac;
+				factors.resize(1);
+			}
+			this->transform = make_shared<Transform<FPP, DEV>>(factors, 1.0, false, cloning_fact);
+			if(cloning_fact) {
+				for(faust_unsigned_int i = 0; i < size; i++)
+					delete factors[i];
+			}
+		}
+
 }

wrapper/python/src/FaustCoreCpp.h

@@ -101,7 +101,7 @@ class FaustCoreCpp
     FaustCoreCpp<FPP,DEV>* slice(unsigned int, unsigned int, unsigned int, unsigned int) const;
     FaustCoreCpp<FPP,DEV>* fancy_idx(unsigned long int* row_ids, unsigned long int
                                   num_rows, unsigned long int* col_ids,
-                                  unsigned long int num_cols);
+                                  unsigned long int num_cols) const;
     bool save_mat_file(const char* filepath) const;
     FaustCoreCpp<FPP,DEV>* optimize_storage(const bool time=false);
     FaustCoreCpp<FPP,DEV>* optimize(const bool transp=false);
@@ -147,6 +147,9 @@ class FaustCoreCppGPU: public FaustCoreCpp<FPP, GPU2>
 		FaustCoreCppGPU<FPP>* horzcat_gpu(FaustCoreCppGPU<FPP>* right) const;
 		FaustCoreCppGPU<FPP>* vertcat_gpu(FaustCoreCppGPU<FPP>* right) const;
 		FaustCoreCppGPU<FPP>* slice_gpu(unsigned int start_row_id, unsigned int end_row_id, unsigned int start_col_id, unsigned int end_col_id) const;
+		FaustCoreCppGPU<FPP>* fancy_idx_gpu(unsigned long int* row_ids, unsigned long int
+                                  num_rows, unsigned long int* col_ids,
+                                  unsigned long int num_cols) const;
 		void device_gpu(char* dev) const;
 		static FaustCoreCppGPU<FPP>* randFaustGPU(unsigned int t,
 				unsigned int min_num_factors, unsigned int max_num_factors,

wrapper/python/src/FaustCoreCpp.hpp

@@ -323,7 +323,7 @@ FaustCoreCpp<FPP,DEV>* FaustCoreCpp<FPP,DEV>::left(const faust_unsigned_int id)
 template<typename FPP, FDevice DEV>
 FaustCoreCpp<FPP,DEV>* FaustCoreCpp<FPP,DEV>::fancy_idx(unsigned long int* row_ids, unsigned long int
         num_rows, unsigned long int* col_ids,
-        unsigned long int num_cols)
+        unsigned long int num_cols) const
 {
     Faust::TransformHelper<FPP,DEV>* th = this->transform->fancy_index(row_ids, num_rows, col_ids, num_cols);
     FaustCoreCpp<FPP,DEV>* core = new FaustCoreCpp<FPP,DEV>(th);

wrapper/python/src/FaustCoreCppGPU.hpp

@@ -148,4 +148,12 @@ FaustCoreCppGPU<FPP>* FaustCoreCppGPU<FPP>::slice_gpu(unsigned int start_row_id,
 {
 	return (FaustCoreCppGPU<FPP>*) FaustCoreCpp<FPP, GPU2>::slice(start_row_id, end_row_id, start_col_id, end_col_id);
 }
+
+template<typename FPP>
+FaustCoreCppGPU<FPP>* FaustCoreCppGPU<FPP>::fancy_idx_gpu(unsigned long int* row_ids, unsigned long int
+        num_rows, unsigned long int* col_ids,
+        unsigned long int num_cols) const
+{
+	return (FaustCoreCppGPU<FPP>*) FaustCoreCpp<FPP, GPU2>::fancy_idx(row_ids, num_rows, col_ids, num_cols);
+}
 #endif

wrapper/python/src/FaustCoreCy.pxd

@@ -87,7 +87,7 @@ cdef extern from "FaustCoreCpp.h":
                                     unsigned int) const
         FaustCoreCpp[FPP]* fancy_idx(unsigned long int* row_ids, unsigned long int
                                   num_rows, unsigned long int* col_ids,
-                                  unsigned long int num_cols)
+                                  unsigned long int num_cols) const
         bool save_mat_file(const char* filepath) const
         FaustCoreCpp[FPP]* optimize_time(const bool transp, const bool inplace,
                                        const int nsamples)

wrapper/python/src/FaustCoreCyGPU.pxd

@@ -38,4 +38,7 @@ cdef extern from "FaustCoreCpp.h":
         void device_gpu(char*) const
         FaustCoreCppGPU[FPP]* slice_gpu(unsigned int, unsigned int, unsigned int,
                                     unsigned int) const
+        FaustCoreCppGPU[FPP]* fancy_idx_gpu(unsigned long int* row_ids, unsigned long int
+                                  num_rows, unsigned long int* col_ids,
+                                  unsigned long int num_cols) const
 

wrapper/python/src/_FaustCorePyGPU.pyx

@@ -671,3 +671,44 @@ cdef class FaustCoreGPU:
 
         core._isReal = self._isReal
         return core
+
+    def fancy_idx(self, indices):
+        cdef unsigned long int[:] row_indices_view
+        cdef unsigned long int[:] col_indices_view
+        core = FaustCoreGPU(core=True)
+        # fancy indexing
+        #convert possible slice (on index 0 or 1 of out_indices) to
+        # an array of indices
+        for i in range(0,2):
+            if(isinstance(indices[i], slice)):
+               indices[i] = list(range(indices[i].start, indices[i].stop,
+                                       indices[i].step))
+        # it's possible that on certain architectures unsigned long int is a
+        # 4-bytes integer
+        # TODO: move to a cross-platform size type (like uint32/64_t from stdlib.h)
+        if(sizeof(unsigned long int) == 8):
+            dtype = np.uint64
+        elif(sizeof(unsigned long int) == 4):
+            dtype = np.uint32
+        row_indices = np.array(indices[0], dtype=dtype)
+        col_indices = np.array(indices[1], dtype=dtype)
+        row_indices_view = row_indices
+        col_indices_view = col_indices
+#        print("FaustCorePy.fancy_idx(), row_indices=", row_indices, " size=",
+#              row_indices.size)
+#        print("FaustCorePy.fancy_idx(), col_indices=", col_indices," size=",
+#              col_indices.size)
+        if(self._isReal):
+            core.core_faust_dbl = \
+            self.core_faust_dbl.fancy_idx_gpu(&row_indices_view[0], row_indices.size,
+                                          &col_indices_view[0], col_indices.size)
+#        else:
+#            core.core_faust_cplx = \
+#            self.core_faust_cplx.fancy_idx(&row_indices_view[0],
+#                                           row_indices.size,
+#                                           &col_indices_view[0],
+#                                           col_indices.size)
+#
+        core._isReal = self._isReal
+        return core
+